Reports of Botched Surgeries Rise After AI Enters the Operating Room

Reports of malfunctions and patient injuries jumped from eight to at least 100 after a sinus surgery navigation device added artificial intelligence, a Reuters investigation found, with some patients suffering strokes and skull damage.

The case tests how regulators, doctors, and manufacturers share responsibility when software that learns from data starts guiding decisions inside a patient's body.

The device in question, the TruDi Navigation System, helps ear, nose, and throat surgeons treat chronic sinusitis, an inflammation of the sinuses.

Acclarent, then a unit of Johnson & Johnson, had sold the system since about 2018 as a navigation tool that tracked surgeons' instruments without any AI.

The company then promoted a 2021 software upgrade to that same device as a major advance, adding a machine-learning algorithm to the navigation system.

The software identifies segments of a patient's anatomy and calculates the shortest valid path between two points the surgeon picks, according to a company post.

At least 10 people suffered injuries between late 2021 and November 2025, and most reports claim the system misinformed surgeons about where their instruments sat inside patients' heads.

One report described cerebrospinal fluid, the clear liquid that cushions the brain, leaking from a patient's nose, and another described a punctured skull base.

In June 2022, Texas patient Erin Ralph suffered a stroke after her surgeon, guided by TruDi, injured a carotid artery, a major vessel feeding the brain, her lawsuit says.

A second patient, Donna Fernihough, suffered a stroke in May 2023 after her carotid artery ruptured during another TruDi-guided operation, according to her suit.

Her suit also alleges Acclarent rushed the technology to market and set a goal of only 80 percent accuracy for some of its new features.

Acclarent denied the allegations in both ongoing cases, and the current owner, Integra LifeSciences, said no credible evidence connects the system's AI to any alleged injury.

Integra LifeSciences added that the reports show only that a TruDi system saw use in surgeries where adverse events occurred.

The injury reports sit in MAUDE, a public FDA database where manufacturers, importers, and hospitals must file suspected device problems.

The FDA cautions that filing a report does not prove a device caused an injury and that the data cannot measure event rates or compare devices.

That limit leaves the core question, whether the AI actually pointed surgeons wrong, to the courts.

The dispute lands amid a boom, with at least 1,357 AI-enabled medical devices now authorized in the United States, double the total through 2022.

Reuters counted at least 1,401 device reports filed since 2021 involving products on the FDA's AI list, including 115 that mention software or algorithm problems.

The reports name dozens of other AI products, including a heart monitor that allegedly overlooked abnormal heartbeats and an ultrasound tool accused of mislabeling fetal body parts.

Most such devices reach the market through the 510(k) pathway, which clears a product if it substantially matches a device already sold.

That comparative standard usually skips fresh clinical trials, the independent proof of safety and effectiveness that full premarket approval demands.

Thin evidence carries measurable risk, according to a JAMA Network Open study that tracked recalls across 903 AI-enabled medical devices.

Devices with no documented clinical studies saw recalls at 7.8 percent, roughly triple the 2.6 percent rate among devices backed by published studies.

Regulators have started asking harder questions, and in September 2025, the FDA requested public comment on how to track AI device performance after deployment.

The agency warned that an algorithm's performance can drift over time as clinical practice, patient populations, and data inputs change.

Its questions cover performance metrics, monitoring triggers, and how human-AI interaction shapes safe use.

The gap between fast authorization and slow real-world checks widens the odds that patients, rather than paperwork, expose a faulty algorithm first.

Both lawsuits continue in Texas courts, leaving judges to sort out a question regulators have not settled: who answers when AI guidance goes wrong.